WaveNet-SF: A hybrid network for retinal disease detection based on wavelet transform in spatial-frequency domain.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinal diseases cause significant vision loss, necessitating early diagnosis.
• Optical Coherence Tomography (OCT) is crucial for retinal imaging but faces challenges like noise and complex lesion features.
• Accurate interpretation of OCT images is vital for effective treatment of retinal conditions.

Purpose of the Study:

• To introduce WaveNet-SF, a novel model for enhanced retinal disease detection using OCT images.
• To improve the accuracy and robustness of automated retinal disease diagnosis.
• To address challenges in OCT image analysis, including noise and feature variability.

Main Methods:

• Developed WaveNet-SF, integrating spatial and frequency-domain learning via wavelet transforms.
• Implemented a Multi-Scale Wavelet Spatial Attention (MSW-SA) module for focused lesion detection.
• Incorporated a High-Frequency Feature Compensation (HFFC) block to preserve details and reduce noise.

Main Results:

• Achieved state-of-the-art (SOTA) classification accuracy of 97.82% on the OCT-C8 dataset.
• Attained SOTA classification accuracy of 99.58% on the OCT2017 dataset.
• Demonstrated superior performance compared to existing methods in retinal disease detection.

Conclusions:

• WaveNet-SF effectively enhances retinal disease detection from OCT images.
• The model's integrated approach overcomes limitations of traditional OCT analysis.
• WaveNet-SF shows significant potential as a tool for diagnosing retinal diseases.